GroundTruth

🚀 GroundTruth

The Physical World API for AI Agents

AI agents can browse the web.
They can call APIs.
But they still can't verify the physical world.

The Missing Capability

Whenever an AI agent needs information only a human can observe, automation stops. GroundTruth removes that limitation.

A single API call lets an agent request real-world verification, receive trusted evidence from a Human Oracle Network, verify it with an AI Notary, and automatically settle payment on X Layer using x402.

How It Works

AI Agent
    │
    ▼
GroundTruth API (MCP)
    │
    ▼
Human Oracle Network
    │
    ▼
Evidence + Freshness Challenge
    │
    ▼
AI Notary
    │
    ▼
X Layer Settlement (x402)

Real-World Use Cases

Why It's Different

GroundTruth is not a crowdsourcing marketplace.

It is infrastructure built specifically for autonomous AI agents.

Every task is:
🤖 AI Generated • 👷 Human Executed • 🧠 AI Verified • ⛓️ On-Chain Settled • 🔌 API Accessible • ⚙️ Fully Automatable

Key Features

Proof

• ✅ Live MCP server

• ✅ Real x402 payment flow

• ✅ Deployed on X Layer Testnet

• ✅ Working AI Notary verification

• ✅ Human Oracle workflow

• ✅ OKX.AI ASP #6282

Business Model

Every successful task creates value for everyone involved.

Vision

The internet became programmable through APIs.
GroundTruth makes the physical world programmable.

GroundTruth is building the infrastructure that gives autonomous AI agents programmable access to the physical world.

Every API connects software to software.
GroundTruth connects software to reality.

Built for #OKXAI Genesis Hackathon
ASP #6282

GroundTruth was built entirely during the hackathon, evolving from an initial prototype into a complete end-to-end infrastructure for AI-powered real-world task execution.

Infrastructure & Payments

• Built an MCP-native service that allows AI agents to create and manage real-world tasks through a single tool call.

• Implemented a real x402 payment flow using Permit2 on X Layer instead of mocked payments.

• Integrated EIP-712 signatures for secure payment authorization.

• Added replay protection and secure settlement validation.

• Deployed the on-chain payroll contract to X Layer Testnet for automated worker payouts.

AI Verification

• Built an AI verification engine for both image and structured-form submissions.

• Added semantic verification to ensure submitted proof matches the requested task.

• Implemented a per-task freshness challenge to prevent replay attacks and reused photos.

• Added explainable verification results so every approval or rejection is transparent.

• Designed the system with a fail-closed architecture, ensuring uncertain verifications never release payment automatically.

Product

• Built the worker marketplace with task claiming, submissions, reputation tracking, and settlement history.

• Added worker leaderboards and proof management.

• Integrated provider fallback, API key rotation, and verification caching for reliability.

• Registered GroundTruth as ASP #6282 on the OKX AI Agent Marketplace.

Throughout the hackathon, GroundTruth was continuously improved through iterative testing and adversarial evaluation, strengthening its payment flow, verification reliability, and production readiness.

Bootstrapped.

GroundTruth is currently an early-stage project developed during the OKX.AI Genesis Hackathon. We are actively seeking design partners, ecosystem collaborations, and strategic funding to expand the protocol into a production-ready decentralized human workforce layer for AI agents.